Thank you for your reply. It's not that simple, though. Just because some core algorithms are standardised and should be compatible doesn't mean their use in different implementations leads to interoperable data. The key point here seems to be that Dovecot just supports SHA-1 with PBKDF2, not SHA-256. So I'm out of luck here. The different formats are no longer relevant then. CRYPT-SHA512 is not anywhere near as secure as PBKDF2. But I've read and learned a lot about secure password hashing in the past 24 hours. My initial point that PBKDF2 is the state of the art has been disproved already. This order seems to be the case [1]: MD5/SHA1 << SHA2 << PBKDF2 < bcrypt < scrypt < Argon2 So I've changed my plans and try to go for Argon2 now. I found support for .NET Core [2] and Python [3]. My original question is kind of obsolete now because I also found another requirement: password rehashing. I'm migrating from an old database that has CRYPT-SHA512 hashes and want to upgrade them to Argon2. This affects multiple services (IMAP, SMTP, FTP, Management UI) so I think I'll better make a central authentication service that has all the passwords and crypto in one place and handles requests from those service daemons. I'm currently investigating how to build such a service and integrate it into the services. Maybe a Unix socket is a good communication channel. Dovecot should be able to query it with a custom Lua script. Haven't looked into the other services yet. That auth service could be built with Python for isolation from other services, high availability and relatively low memory footprint. It connects to the database, reads and updates the hashes and does all the crypto for its clients. Any suggestions about how to do that? Yves [1] https://cryptobook.nakov.com/mac-and-key-derivation/argon2 [2] https://github.com/tabrath/libsodium-core [3] https://passlib.readthedocs.io/en/stable/lib/passlib.hash.argon2.html -------- Urspr?ngliche Nachricht -------- Von: Aki Tuomi <aki.tuomi at open-xchange.com> Gesendet: Sonntag, 30. August 2020, 16:33 MESZ Betreff: PBKDF2 password hashing as in ASP.NET Core Hi! The PBKDF2 algorithm is standard and should be compatible with ASP.NET Core. The salt parameter is 16 symbols from the salt character set ./0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz followed by number of rounds hash is hex encoded 160-bit value which comes out of the PBKDF2 function with SHA1. Dovecot does not currently have support for PBKDF2-SHA256, only PBKDF2-SHA1. You could use CRYPT-SHA512 instead which is probably just as good? Aki
In case you are interested, https://wiki.dovecot.org/HowTo/ConvertPasswordSchemes By the way, I am bit sceptical that CRYPT-SHA512 is less secure than PBKDF2. CRYPT-SHA512 is not "just" SHA512(salt||password), it does at least 1000 rounds of hashing in similar way as PBKDF2 does. So, what is your reasoning for claiming that PBKDF2 is much secure than CRYPT-SHA512? Also, if you look at hashcat cracking speeds, you'll see that the speed of cracking is slower for CRYPT-SHA512 than for PBKDF2-SHA512. See https://github.com/siseci/hashcat-benchmark-comparison/blob/master/1x%20GTX%201080%20TI%20hashcat%20benchmark.txt Aki> On 30/08/2020 19:54 Yves Goergen <nospam.list at unclassified.de> wrote: > > > Thank you for your reply. > > It's not that simple, though. Just because some core algorithms are > standardised and should be compatible doesn't mean their use in > different implementations leads to interoperable data. The key point > here seems to be that Dovecot just supports SHA-1 with PBKDF2, not > SHA-256. So I'm out of luck here. The different formats are no longer > relevant then. > > CRYPT-SHA512 is not anywhere near as secure as PBKDF2. > > But I've read and learned a lot about secure password hashing in the > past 24 hours. My initial point that PBKDF2 is the state of the art has > been disproved already. This order seems to be the case [1]: > > MD5/SHA1 << SHA2 << PBKDF2 < bcrypt < scrypt < Argon2 > > So I've changed my plans and try to go for Argon2 now. I found support > for .NET Core [2] and Python [3]. > > My original question is kind of obsolete now because I also found > another requirement: password rehashing. I'm migrating from an old > database that has CRYPT-SHA512 hashes and want to upgrade them to > Argon2. This affects multiple services (IMAP, SMTP, FTP, Management UI) > so I think I'll better make a central authentication service that has > all the passwords and crypto in one place and handles requests from > those service daemons. > > I'm currently investigating how to build such a service and integrate it > into the services. Maybe a Unix socket is a good communication channel. > Dovecot should be able to query it with a custom Lua script. Haven't > looked into the other services yet. That auth service could be built > with Python for isolation from other services, high availability and > relatively low memory footprint. It connects to the database, reads and > updates the hashes and does all the crypto for its clients. > > Any suggestions about how to do that? > > Yves > > > [1] https://cryptobook.nakov.com/mac-and-key-derivation/argon2 > [2] https://github.com/tabrath/libsodium-core > [3] https://passlib.readthedocs.io/en/stable/lib/passlib.hash.argon2.html > > > > -------- Urspr?ngliche Nachricht -------- > Von: Aki Tuomi <aki.tuomi at open-xchange.com> > Gesendet: Sonntag, 30. August 2020, 16:33 MESZ > Betreff: PBKDF2 password hashing as in ASP.NET Core > > > Hi! > > The PBKDF2 algorithm is standard and should be compatible with ASP.NET Core. > > The salt parameter is 16 symbols from the salt character set > > ./0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz > > followed by number of rounds > > hash is hex encoded 160-bit value which comes out of the PBKDF2 function > with SHA1. > > Dovecot does not currently have support for PBKDF2-SHA256, only > PBKDF2-SHA1. You could use CRYPT-SHA512 instead which is probably just > as good? > > Aki
I'm not an expert at this, I just read other sources that make suggestions. I have no strong comparison of SHA2 and PBKDF2. So my "<<" may be exaggerated. The most important points today seem to be resilience against GPU and ASIC attacks. Those devices have only little memory to work with. SHA2 is said to be designed to use very little memory which only leaves the CPU load as defence. This can be compensated by massive parallelisation, as in GPUs. Neither SHA2 nor PBKDF2 should be resilient against GPU attacks. This is where the other algorithms are better because they introduce bigger memory requirements. I'll probably look into PAM authentication and see if I can get my own module into there somehow. This should be the most versatile method to provide secure authentication for all services. At least Dovecot, Exim and ProFTPd support PAM. -------- Urspr?ngliche Nachricht -------- Von: Aki Tuomi <aki.tuomi at open-xchange.com> Gesendet: Sonntag, 30. August 2020, 19:27 MESZ Betreff: PBKDF2 password hashing as in ASP.NET Core In case you are interested, https://wiki.dovecot.org/HowTo/ConvertPasswordSchemes By the way, I am bit sceptical that CRYPT-SHA512 is less secure than PBKDF2. CRYPT-SHA512 is not "just" SHA512(salt||password), it does at least 1000 rounds of hashing in similar way as PBKDF2 does. So, what is your reasoning for claiming that PBKDF2 is much secure than CRYPT-SHA512? Also, if you look at hashcat cracking speeds, you'll see that the speed of cracking is slower for CRYPT-SHA512 than for PBKDF2-SHA512. See https://github.com/siseci/hashcat-benchmark-comparison/blob/master/1x%20GTX%201080%20TI%20hashcat%20benchmark.txt Aki